Analysis · curated 19 Jul 2026
Physical AI Security: A Threat Model for Edge Devices
First reported exein.io
Coverage timeline
Single-source analysis — first reported, latest, and curated coincide.
Why it matters
Physical AI devices turn indirect prompt injection into an attack that arrives as light through a lens with no packet or CVE, giving defenders a framework to reason about jailbreaks that can move robot arms or exfiltrate camera footage at fleet scale.
Exein's blog post "Physical AI Security: A Threat Model for Edge Devices" argues that on-device AI (cameras, robots, drones running vision-language models and LLMs locally) introduces risks classic embedded threat models miss: every sensor becomes an instruction channel enabling physical-world prompt injection (e.g. text on a sign in front of a camera), probabilistic behavior that cannot be patched like a CVE, and unattended failures that act on the physical world via actuators. It proposes modeling the agent loop rather than individual components.